AGI does not replace "everything". It might replace most of the work that someone can do behind a desk, but there are a lot of jobs that involve going out there and working with reality outside of the computer.
AGI as defined these days is typically “can perform at competent human level on all knowledge work tasks” so somewhat tautologically it does threaten to substitute for all these jobs.
It’s a good thing to keep in mind that plumbers are a thing, my personal take is if you automated all the knowledge work then physical/robot automation would swiftly follow for the blue-collar jobs: robots are software-limited right now, and as Baumol’s Cost Disease sets in, physical labor would become more expensive so there would be increased incentive to solve the remaining hardware limitations.
I don't think that robots are software-limited in all domains... And even if AI became a superhuman software dev, it wouldn't drive the cost of software development to zero.
Humanoid robotics (ie replacing the majority of workers) is highly software-limited right now.
Here’s a napkin-sketch proof: for many decades we have had hardware that is capable of dextrously automating specific tasks (eg car manufacture) but the limitation is the control loop; you have to hire a specialist to write g-code or whatever, it’s difficult to adapt to hardware variance (slop, wear, etc) let alone adjust the task to new requirements.
If you look at the current “robot butler” hardware startups they are working on: 1) making hardware affordable, 2) inventing the required software.
Nothing in my post suggested costs go to zero. In the AGI scenario you assume software costs halve every N years, which means more software is written, and timelines for valuable projects get dramatically compressed.
Also, presumably if you have AGI you can have it address a physical problem at a higher level of abstraction. "Design a device to make any water heater installable by a single person in 20 minutes" would result in a complex system that would make a lot of blue collar labor redundant (the last water heater I had installed took 3 guys over an hour to complete).
It would not even necessarily result in a human-like robot - just some device that can move the water heater around and assist with the process of disconnecting the old one and installing the new one.
There are already plenty of supply chain problems in the AI industry, but the supply chain limitations to robotics are even higher. You can't snap your fingers and increase robotics production tenfold or a hundredfold without a lot of supply chain improvements that will take a long time. I'd say anywhere between twenty and fifty years.
"Everything" might be hyperbole but a huge percentage of the workforce in my country is office/desk based. Included in that % is a lot of the middleclass and stepping stone jobs to get out of manual work.
If AI kills the middle and transitional roles i anticipate anarchy.
It isn't really general intelligence if it doesn't compete, or outcompete us, at a large enough portion of knowledge work jobs that we are forced out of the labor market.
Also don't forget that plenty of knowledge work is focused on automating manual labor. If AGI is a thing, it will eventually be used to also outcompete us on physical work too.
People like to point to plumbers as an example of a safe(r) job, and it is. But automating plumbing tasks is most difficult because the entire industry is designed to be installed by humans. Without that constraint it would likely be much easier to design plumbing systems and robots to install and maintain them more efficiently than what we have today with human-optimized plumbing.
I think about all the actual physical work that goes into building a functioning supply chain, and that it will take a lot more than "figuring out" to manifest such a physical form.
